Why don't eligible firms claim hiring subsidies? The role of job duration.

I. INTRODUCTION

For over 30 yr, the U.S. government has encouraged the hiring of disadvantaged workers by allowing firms with qualified workers to claim federal tax credits. The Work Opportunity Tax Credit (WOTC) and Welfare-to-Work Tax Credit (WtW) are available to firms that hire certain workers from a number of target groups, such as welfare recipients, young food stamp recipients, young workers from economically depressed areas, and ex-felons. Despite substantial labor market participation by these groups, only a small percentage of firms claim the subsidies for which they qualify. For example, Wisconsin microdata suggest that employers left an estimated $8 million unclaimed for qualified welfare and food stamp recipients in 2001; in the same year, employers claimed less than $2 million for workers in these target groups. Because Wisconsin only accounts for about 2% of the U.S. population (and 2% of all WOTC/WtW certifications), unclaimed benefits nationwide are likely to be worth over $400 million per year. Moreover, low participation in employer subsidy programs is not a new phenomenon but has persisted for decades in a variety of similar programs.

Why do firms leave these subsidies unclaimed? Possible explanations for this puzzle include lack of information, discomfort with involvement in government programs, or low benefits from participation relative to administrative costs. In this study, I examine the merits of the last explanation. In particular, I investigate the potential role of worker "hours requirements," which tightly link subsidy rates to the job duration of qualified employees. Such requirements have been a standard feature of targeted employer tax credit programs since the 1970s. In the period I study (1999-2002), a firm qualifies for the WOTC only if an eligible worker remains employed for at least 120 h, and the subsidy rate increases if the worker exceeds 400 h. The WtW credit requires a minimum of 400 h of work; however, workers in this category can qualify for the WOTC if they do not meet this hours requirement. A firm must apply for the tax credits at the time it hires a qualified worker, even though the firm will only be permitted to claim a subsidy if the worker's hours exceed the threshold. This combination of up-front costs and uncertain benefits might discourage some firms from participating. Understanding the potential role of hours requirements in the participation decision may provide needed insight into firms' continually low interest in employer subsidies.

In this article, I examine the relationship between a firm's WOTC/WtW participation and the distribution of its eligible workers' job durations, which is a key determinant of potential benefits. I use a unique set of Wisconsin administrative data, which links the state's WOTC/WtW job records (containing hourly wage data) to their corresponding unemployment insurance records (containing total earnings data). This allows me to conduct the first examination, to my knowledge, of the distribution of job durations (in hours) for subsidy-certified workers. (1) I then compare this distribution with an estimate of the hours distribution for other WOTC/WtW-eligible workers whose employers did not apply for the credits. A reasonable explanation for these firms' nonparticipation may be that they would have gained little from the subsidy program. If firms form expectations based on their typical hours distribution, then I expect to see higher participation among firms with stronger distributions, that is, with more workers in the higher subsidy brackets. I estimate the average fixed cost of firm participation in WOTC and then, using a probit model in which I control for relevant covariates, I analyze the relationship between firms' worker-hour distributions and their participation in the WOTC/WtW.

Although this relationship is interesting, it does not necessarily imply a causal effect of hours on participation if, as theory would predict, participating firms maximize WOTC/WtW benefits by intentionally increasing workers' job duration (after the participation decision has already been made). Fortunately, the discontinuities in WOTC/WtW subsidy levels allow me to investigate this possible "reverse causality." I do this by examining firms' responses to the particularly strong incentive to increase job durations near the hours thresholds, where benefits from influencing job duration would be greatest. Small increases in job duration near the thresholds can provide net benefits to firms, as the change in the subsidy rate more than compensates for the wage costs of a few additional hours of work. Professional tax credit consultants recognize this; for example, one consultant provides firms with "Credit Maximizer" reports that enable companies to "closely monitor all employees eligible for the WOTC that are less than 10 h away from qualifying for the tax credit." (2) If participating firms can influence job duration on the margin, the hours distribution should contain a dip for hours levels just under each threshold, and a spike just above, indicating that a firm was able to extend the job duration of some workers to take advantage of the higher rate. Examining these distributions allows me to properly infer causality in the analysis of hours distributions and WOTC/WtW participation.

II. THE WORK OPPORTUNITY AND WELFARE-TO-WORK TAX CREDITS

The WOTC was introduced in 1996 and can be claimed by any for-profit firms that hire workers from a variety of targeted disadvantaged groups, such as workers with at least 9 mo of recent welfare receipt and young workers (18-24) with at least 6 mo of recent food stamp receipt. The WtW was introduced in 1997 and applies only to workers with at least 18 mo of recent welfare receipt. Both credits have been reauthorized multiple times and have very recently been combined (although the following discussion will refer to the distinct policies in place during the period under study). (3) For a firm to claim the WOTC, which is available for 1 yr, an eligible employee must work at least 120 h to qualify the firm for a 25% subsidy and at least 400 h for a 40% subsidy. The WtW requires 400 h of work to obtain a 35% subsidy, and allows a second year of subsidy at a 50% rate. These tax credits are subject to a per-worker cap, set at $2,400 per WOTC worker (for 1 yr only) and $8,500 per WtW worker (total for 2 yr). (4) Those who quality for the WtW automatically qualify for the WOTC as well, because it requires less welfare receipt than the WtW. Firms can apply for certification for both credits when they hire a long-term welfare recipient but are permitted to claim only one (which they can choose ex post based on whichever is more advantageous). The WOTC is usually more generous than the WtW in the first year, and only a small fraction of workers continue work into a second year, so the WtW is claimed much less frequently. (5) For simplicity, I refer to the pair of programs as "WOTC" in cases where the distinction between them is unimportant.

The WOTC is essentially a reformed version of the Targeted Jobs Tax Credit (TJTC). The TJTC began in 1978 and was allowed to expire in 1994 after a report from the Office of the Inspector General (OIG) indicated little success in improving labor market outcomes for the targeted groups (U.S. Department of Labor. OIG 1994). A key distinction between the TJTC and the WOTC is a difference in hours requirements. When it expired, the TJTC had a single 120-h threshold and a 40% subsidy rate. When the WOTC was introduced in 1996, it maintained the single-rate system but reduced the rate to 35% and required at least 400 h of work (in an effort to encourage longer job tenure). (6) However, in the reauthorization process just a year later, a new "lower level" (25%) subsidy bracket was introduced for workers who were employed for at least 120 h, and the subsidy rate for the higher bracket (400 h +) was increased to its current level of 40%. Figure 1 illustrates this schedule of potential benefits to a firm in the case of a worker earning $6 per hour (a typical wage for a Wisconsin WOTC worker). The incentives created by this structure are a key issue in this study and will be discussed in the analysis of firms' participation decisions. The rationale for the change to a two-tier structure was stated clearly by Senator Lincoln Chafee:

This is necessary because many employers were finding it difficult to retain these employees for ... 400 hours, and as a result were losing the benefits of the tax credit. In many cases the employers were spending the money to train the employees only to have them leave shortly thereafter for higher paying jobs. Without some reward for their efforts, employers were simply dropping their programs. (143 Congressional Record S 8415)

Even though policy makers had intentionally imposed hours requirements to encourage long-lasting jobs, it is noteworthy that early in the program there was already concern that hours requirements may discourage firms from participating. However, no research has been undertaken to investigate this issue.

III. WOTC PARTICIPATION AND ITS POTENTIAL BENEFITS TO FIRMS

Some particular features of the administrative process for claiming the WOTC may affect firms' willingness to participate. First, the application is quite simple and is the same for both tax credits. This paperwork is less burdensome than that under the TJTC--it is limited to just two one-page forms. First, firms have job applicants or new hires fill out a "Pre-Screening Notice" (Internal Revenue Service [IRS] form 8850) that requires the worker to fill in basic information (name, etc.) and to check a box if he/she may qualify for one of the listed target groups. If (and only if) the worker checks one of the boxes, a second one-page form (Employment and Training Administration [ETA] 9061) must be submitted to the state with additional documentation. In many cases, state offices or firms' tax consultants handle the documentation. (7) This simplification of paperwork was intended to promote higher levels of participation than those under the TJTC. Second, the program requires that the initial eligibility assessment be filled out on or before the date of hire. This, too, is a reaction to concerns about the TJTC, through which firms could apply for subsidies for any of their currently eligible workers (rather than new hires only), resulting in little effect on hiring (Lorenz 1995). Firms submit their paperwork to the state's relevant employment division and later receive a certification letter for each approved worker. Thus, while the paperwork is not particularly burdensome, firms must pay their costs of participation "up front"--both by learning how to implement the program (fixed costs) and by filling out each worker's paperwork at the time of hire (marginal costs). The benefits of participation are determined later, via the job duration and earnings of certified workers. Firms claim the tax credit on their federal tax forms, where they are only asked to report the total firm value of qualified worker earnings at each subsidy rate.

[FIGURE 1 OMITTED]

The evidence on participation in the WOTC indicates that the vast majority of eligible workers are not claimed by their employers for the subsidies. Hamersma (2003) estimates the number of WOTC-eligible workers in the food stamp and welfare target groups nationally (counting only those who obtained a job while eligible) and compares these estimates with the number of WOTC certifications granted for these groups. Participation rates in 1999 are estimated to be between 1% and 17% for food stamp youth and between 10% and 32% for welfare recipients. These represented increases from the first year of available data examined in the paper (1997), and perhaps increases relative to the TJTC, for which estimates ranged from less than 5% to about 10% (see Christensen 1984; Katz 1998; O'Neill 1982). However, the estimated levels remain inexplicably low.

A variety of explanations have been proposed for employers' lack of participation in the WOTC and similar programs. In studies of the TJTC, researchers suggested that information costs were a main reason for firms' lack of participation (Bishop and Kang 1991). Given the availability of current program descriptions and forms on the internet, this reason is less compelling now than in the past (although a completely unaware firm may not even know that the information was easily available). Other researchers have emphasized the role of worker stigma. Burtless (1985) found that workers who were instructed to present TJTC vouchers to prospective employers experienced less success in finding employment than similar workers without the vouchers. Dickert-Conlin and Holtz-Eakin (2000) presented a theoretical framework for modeling such employer-subsidy stigma. However, under the WOTC, workers do not need to initiate discussion of their eligibility as they did under some forms of the TJTC. It is possible that some workers choose not to fill out the WOTC form enclosed in their job application (or to fill it out dishonestly) due to stigma, but it is not clear whether this concern would induce firms to decide not to participate at all in the program unless this practice was so widespread that firms rarely had an honest, eligible applicant. Finally, some have argued that firms are simply uncomfortable with involvement in government programs, perhaps because they believe they may become more susceptible to audits or other government intervention. This discomfort would have to be significant to generate such low interest in a program that can provide substantial tax relief.

One plausible explanation for low participation that has not been sufficiently explored is that participation is unattractive because benefits of participation are both uncertain and limited due to hours requirements. Although the potential benefit of participation is $2,400 per WOTC-certified worker (and $8,500 over 2 yr per WtW-certified worker), the maximum benefit is not necessarily relevant for firms considering participation. On the basis of the total federal tax costs of the program relative to the number of certified workers, Hamersma (2003) estimated average WOTC returns of about $1,000 per worker in the year 2000. Some employers seem to be aware of the potential for limited returns. In a survey of 101 temporary help services firms, Hamersma and Heinrich (2008) found that 26 of the 73 firms that had heard of the WOTC or WtW did not participate. Several indicated that they did not have enough eligible workers, and a few indicated that workers' job duration was too short to make participation worthwhile. This suggests an underlying fixed cost of participation as well as an awareness of possibly limited benefits. The remainder of this article investigates how the limits on subsidies for short-duration jobs may affect firms' WOTC participation decisions.

IV. DESCRIPTION AND SUMMARY OF DATA

There is no direct source of job duration data for any past or current employer subsidy participants. This is an important reason that there have not been any studies on their hours distributions. In the case of the WOTC, the Department of Labor (DOL) administers the certification for the programs, handling employer applications and certifying workers' disadvantaged status, but it does not collect data on job duration following certification. Instead, the IRS is charged with enforcing the hours requirements when firms claim the tax credits. The form that employers submit to the IRS is very brief, reporting only the total dollar value for which they are claiming each subsidy rate. (8) It follows that even merged DOL and IRS data would not provide information on individuals' job duration, despite the important role it plays in determining benefits.

I have assembled unique merged data from the state of Wisconsin that allow me to examine the hours worked by subsidy-certified individuals for the first time. The state WOTC database for late 1998 through early 2003 supplies information from all WOTC and WtW applications, including each worker's Social Security Number (SSN), target group (e.g., welfare, food stamps), occupation, employer identification number, tax consultant identifier (if any), job start date, certification date (if application is approved), and starting wage (reported in one-dollar-wide brackets). I merge these data to the state's unemployment insurance (UI) employment records for 1995-2004 by individual SSN. The UI records provide quarterly earnings data, industry, and employer identification numbers. By matching the employer identification numbers with the WOTC records, I can identify the specific WOTC job in the UI data. I then calculate the job duration at this firm (in hours) by dividing total UI earnings by the starting wage given in the WOTC record. (9)

Summary statistics for the Wisconsin WOTC-certified population are provided in Table 1. Most workers are either welfare recipients or food stamp recipients. (10) Note that although there are many WtW certifications, employers may ultimately claim the WOTC for these workers instead if it is more advantageous (which is often the case), so I treat these long-term welfare recipients as regular WOTC certifications for purposes of my analysis. Table 1 reports that over 60% of all certified workers earn less than $7.00 per hour; these low wages correspond to a high concentration of their jobs in occupations like "Service" or "Clerical and Sales." These jobs are often in the "Retail Trade" or "Accommodation and Food Service" industries, which have been vocal supporters of the WOTC program nationally. All these statistics suggest that disadvantaged workers in Wisconsin are not particularly different from those elsewhere (Richer, Savner, and Greenberg 2001).

The distribution of WOTC workers across different subsidy rates is of primary interest for this study. The estimated distribution from my data is shown in Figure 2. Workers are categorized based on their estimated hours of work, which I obtain by dividing their total first-year earnings at the job by their starting wage, which is coded as the midpoint of their reported wage category. (11) In doing this, I effectively make two assumptions: that the midpoint of the reported wage category is a reasonable approximation of the wage, and that the starting wage continues to be the wage throughout employment. The first assumption should not have strong effects on my findings, as even with a small amount of measurement error in wages I am unlikely to misclassify many people into the wrong subsidy bracket (in other words, only those on the margin between two brackets could be affected). As long as people are classified into the correct brackets, their estimated subsidy values will be accurate because those calculations--discussed later in the article--use total earnings (not wages) that are directly reported in the UI data. The second assumption--that the starting wage is a good measure of the wage throughout employment--is likely to hold in most cases due to the short job durations observed in my sample. (12) However, for workers who obtained raises it will overestimate job duration. This will only overstate the value of these workers' WOTC certification if it is mismeasured enough to place them in the wrong subsidy bracket.

[FIGURE 2 OMITTED]

I find that over one-third of certified workers have less than 120 h of employment, failing to qualify their employer for the WOTC. Another 29% of workers qualify only for the lower subsidy rate of 25% of wages because they do not meet the 400-h threshold. The large number of workers in these low-duration categories is consistent with data reported by Gunderson and Hotchkiss (2007), who examined a particular occupation within one large firm and found that about half of WOTC hires who separated within their 2-yr sample period separated within 30 d. (13) Only a little over one-third of workers in my sample reach the highest subsidy rate (working 400 h or more), and only about half of them qualify for the maximum subsidy of $2,400. The wide variance in workers' hours confirms that job duration is a key determinant of subsidy returns. A firm's ability to influence or predict workers' job duration may affect participation in the WOTC.

In order to examine firms' participation decisions, I also need data on job duration of WOTC-eligible workers whose firms do not claim the credits for which they qualify. I use Wisconsin's public assistance database, called Client Assistance for Re-employment and Economic Support (CARES), to examine the universe of workers who received welfare or food stamp benefits between 1998 and 2001. These workers' records allow me to identify their WOTC eligibility via welfare or food stamps during July 1999 to December 2001. I focus my analysis on these groups to the exclusion of the other target groups (such as ex-felons and low-income veterans) because data are not available for other groups. I limit the CARES data to the workers I find to be WOTC-eligible and then link these workers' records to the WOTC files to determine which of them were actually certified. Note that this sample does not include every certified worker, because some were certified due to their membership in other target groups and some were certified through the welfare or food stamp target groups but do not appear eligible according to the CARES records available to me. (14) (Further discussion of the data sources can be found in the Appendix, and details of sample construction are available from the author upon request.) By using data only for those who appear eligible via welfare or food stamp receipt, I ensure that both the certified and uncertified workers in the sample have similar public assistance records. My sample of eligible workers contains 35,004 person-jobs, of which 2,622 were WOTC certified.

V. JOB DURATION ATWOTC-PARTICIPATING FIRMS AND NONPARTICIPATING FIRMS

If eligible nonparticipants in the WOTC are firms with relatively short worker-job duration, this could help explain the puzzle of low participation rates. However, an association between job durations and participation may also reflect differing firm behavior after the participation decision. I test for a relationship between job duration and firms' participation in the WOTC, and then investigate the direction of causality.

There is an important limitation of the wage data for eligible but uncertified workers: because most of them do not have WOTC records to provide wage data, only their total earnings from the UI data are available. In order to estimate job duration, I need to impute wages for this group. I do this by first estimating a standard log wage regression using demographic and job characteristics of the subsample for which wages are available. Note that although the wage data are measured with error (because they are imputed from categories), this error is in the dependent variable and so it is unlikely to bias the results, although it will make estimates less efficient. This regression has strong explanatory power for a wage regression (adjusted [R.sup.2] = .404), probably due to the homogeneity of the sample. I use the estimated coefficients from this regression (reported in Appendix Table A1) to predict wages for the portion of the sample without wage data. (15)

Using these coefficients to impute wages is appropriate if there are no systematic differences between the groups with and without wage data. However, most of those with wage data were WOTC certified (some had only applications), and most of those without wage data were not WOTC certified. If there is a WOTC wage premium due to the subsidy passing through to WOTC workers, the imputed wages may be too high for the (mostly uncertified) group with missing wage data. For this reason, I have repeated all the analysis that follows using imputed wages that are 10% lower than the regression predictions. (16) The results of my job duration analysis do not change substantially, primarily because relatively few of the workers' hours are near the WOTC thresholds, so the categorization of WOTC subsidy rates is quite robust.

These imputations make it possible to compare the job duration distributions of WOTC-participating and nonparticipating firms. It is easy to define a nonparticipating firm as a firm that did not apply for the WOTC during my period of interest. Defining a "participating" firm is less obvious, because some firms participate only for a small fraction of the workers who are eligible at their firm during this time period. This is surprising given the anecdotal evidence that participating firms often include WOTC forms in every job application or hiring packet they distribute. (17) The reasons for this may include changes in the firm's WOTC policies over time, differences in participation across different physical branch locations within a company, or selective WOTC application (i.e., applications are submitted only for certain workers). (18) The distribution of the fraction of eligible workers claimed (after the firm began participating in the program) is reported in Appendix Figure A 1. The comparisons below define "participating" firms as those that applied for WOTC for at least 20% of their eligible workers (in my sample) after they began participating in the program. (19) I drop observations from firms that participate if they do so for fewer than 20% of their eligible employees. I also drop firms with fewer than ten eligible workers in my sample to focus attention on firms that could have nontrivial gains from participation. I later check the robustness of the results to these choices.

On the basis of the estimated hours distributions shown in Figure 3, firms that participate in the WOTC tend to have workers with longer job durations. It is particularly noticeable that a substantially higher fraction of workers at nonparticipating firms have job durations of less than 120 h, the threshold for WOTC benefits. The median duration is 138 h in participating firms and only 95 h in nonparticipating firms. A Wilcoxon rank-sum test strongly rejects the hypothesis of equality of the two distributions.

Some of the difference in hours distributions may be driven by trends in particular industries. Over 93% of the workers in my sample work in one of the four industries: "Accommodation and Food Service" (18%), "Health Care and Social Assistance" (18%), "Retail Trade" (20%), or "Administrative and Support and Waste Management" (38%). This last category contains mostly (80%) Temporary Help Services (THS) workers. The first industry has fairly minimal differences in the hours distributions between workers in participating and nonparticipating firms, and the "Health Care" category is in the opposite direction from that expected (with about 20% longer job durations in nonparticipating firms). The overall pattern appears to be driven by the "Retail Trade" and "Administrative" (mostly THS) workers. In "Retail Trade," median tenure in nonparticipating firms is 115 h versus 159 h in participating firms. In the "Administrative" category, median tenure is extremely low among both participating and nonparticipating firms, reflecting the nature of THS employment: nonparticipating firms have median tenure of 45 h (with 70% failing to exceed 120 h of work), whereas participating firms have median tenure of 82 h (with 58% failing to exceed 120 h of work). These differences across industries underscore the importance of considering industry differences when assessing job duration distributions and the participation decision.

[FIGURE 3 OMITTED]

VI. ESTIMATION: HOW ARE WORKER HOURS RELATED TO FIRM WOTC PARTICIPATION?

A firm's job duration distribution may affect its participation decision through its effect on potential WOTC tax credits. I use workers' earnings and estimated job duration to estimate the tax credit for which each worker qualifies. By adding these up by firm, I can examine each firm's potential tax credits and then consider how participation might depend on this value. I first obtain an estimate of the fixed costs of participation, and then follow this with an analysis of how workers' hours distributions are related to firm participation.

There is strong evidence that firms' potential returns are systematically correlated with participation in Table 2, which categorizes firms by their potential tax credits and reports the fraction of firms that participate in the WOTC. It is clear that firms with larger gains are more likely to participate. Firms' participation does not appear to respond to potential credits until their total value is over $20,000, after which participation increases with potential benefits, but never gets above 60%. (20) This finding points toward the relevance of perceived fixed costs of participation; lower participation when benefits are low (relative to costs) is consistent with the comparative static predictions of typical theories of firm behavior. (21) These fixed costs include learning about the program and establishing a system for submitting applications. The vast majority of WOTC-claiming firms in Wisconsin use tax consultants, and these consultants may also charge a fixed fee; however, anecdotal evidence on consultant fees suggests that they are typically a commission payment based on the value of credits claimed. (22) Although the administrative fixed costs seem fairly limited based on the application process, it appears that many firms only consider participating if these costs can be spread over a large number of qualified workers.

I estimate fixed costs by using a simple model of firm participation in which firm i compares its expected tax benefits, [B.sub.i], with its fixed costs of participating, [C.sub.i], to decide whether participation is worthwhile. Formally, the decision rule is:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This can be modeled with a probit estimator of the form:

(2) [participation.sub.i] = [PHI](C + [BETA] x [B.sub.i] + [[epsilon].sub.i]),

where the error term allows for the variation in fixed costs, and C represents the average fixed costs of firm participation in WOTC. The prediction is that the estimate of C, because it is a cost, will be negative and that the estimate of [beta] will be positive.

The resulting estimates are provided in Table 3. The signs on the parameters are as expected, and both are statistically significant. There are a couple of ways of interpreting the role of fixed costs on the basis of these results. First, consider the estimated effect of fixed costs on the average probability of participation. The average predicted probability of participation is 18.3%, but with fixed costs removed this average would increase to 58.6% (near the maximum participation rate in the sample, on the basis of Table 2). This suggests that, on average, fixed costs reduce the probability of participation by about 40% points. Second, consider the actual average dollar value of fixed costs. One can divide the intercept estimate by the coefficient estimate on benefits to determine how many dollars of expected benefits would be required to negate the average fixed costs. I find that removing fixed costs would be equivalent to an increase in the level of benefits of about $145,000. (23) By both measures, fixed costs are large and influential. However, one should keep in mind that these are estimates of average fixed costs; on the basis of the wide distribution of participation rates in Table 2, it is clear some firms experience much lower fixed costs whereas others much higher. However, despite wide variation in fixed costs, firms do have a common pattern of higher participation when potential tax credits are higher, all else equal.

It is possible that this strong relationship between potential tax credits and participation in the WOTC would exist even without an hours requirement that differentiates subsidy rates; for instance, higher potential tax credits at participating firms may simply be due to a larger number of eligible workers or higher wages. In order to disentangle these effects, I estimate a participation equation in which I do not control for the potential tax credit itself, but instead include the number of eligible workers at each firm, the average wage paid by the firm to these eligible workers, and a set of variables representing the distribution of workers across the 4-h-threshold categories (percent less than 120 h, 120-399 h, over 400 h but below maximum subsidy, and maximum subsidy). This combination controls for the multiple factors influencing potential tax credits, with the advantage of allowing estimates of the distinct importance of the job duration of WOTC-eligible workers in the participation decisions of their employers. I estimate these effects of job duration on participation using a probit equation in which I also control for industry, region, and the dispersion of the firm (proxied by the number of counties in which it has sample employees). This specification allows for differences in information, administration, and interest in employer subsidies across industries.

The results, shown in Table 4, indicate a strong relationship between a firm's job duration distribution and its participation in the WOTC. The three variables representing the percent of workers with each job duration (120-399 h, 400 h or more but not at maximum subsidy, or maximum subsidy) are jointly significant with a p value of .002. This suggests that a firm's probability of participating is higher when its hours distribution is more heavily concentrated in the WOTC-qualifying range (relative to the default of less than 120 h of work). The estimated marginal effect is largest and most precisely estimated for the percent of workers with more than 400 h of work but below the maximum subsidy: for every 2 percentage-point increase in this variable (relative to the less than 120 h category), the probability of firms' participation in the WOTC increases by just over 1 percentage point. (24) Because the raw participation rate of firms in the sample is only 18.2%, this represents a meaningful change. The number of eligible workers is also important, as the estimates suggest that the addition of 100 eligible workers is associated with a 4 percentage-point increase in a firm's probability of participating in the WOTC. The average wage is positively associated with participation as well. Thus, it appears that all of the contributors to potential tax credits--including the number of eligible workers, their wages, and the distribution of their hours--are positively related to firms' participation.

My findings of a positive relationship between hours distributions and firm participation in the WOTC could be sensitive to a number of decisions I made in my sample choice. To address this, I examined a variety of alternative specifications. First, I tried including all firms with five or more eligible workers (instead of ten) and, separately, only those with 15 or more eligible workers. I considered my participation definition as well, testing thresholds of "any participation" (>0%), 10%, or 30% (instead of 20%). Finally, I applied the participation threshold on the basis of a firm's fraction claimed over the whole sample period (rather than the fraction claimed after the firm started participating). Examining every combination of these choices led to 24 distinct sets of estimates. Of these, all specifications had a statistically significant positive coefficient on at least one of the three hours distribution indicators, most (18) specifications had statistically significant positive coefficients on at least two of these indicators, and no specification had a statistically significant negative coefficient on any of them. (25) This consistency is remarkable given the small size of the sample (less than 400 firms) in some specifications. Thus, it appears the specific choices of sample are not driving my finding of a positive relationship between hours and participation.

VII. REVERSE CAUSALITY: DO FIRMS INCREASE THEIR WOTC BENEFITS BY INFLUENCING JOB DURATION?

To conclude that different job duration distributions have a causal impact on firms' participation decisions, I need to establish that firms' job duration distributions are exogenous to participation. This is not at all obvious, because one might expect WOTC-participating firms to adjust their hours distributions in response to the program's incentives. The established difference in hours distributions across participating and nonparticipating firms would then be a result of the program itself ex post, and not necessarily a factor in the participation decision ex ante. In many situations, it would not be possible to distinguish between these possibilities; however, the discontinuity created by the subsidy brackets provides a unique opportunity to investigate this issue.

Because WOTC-certified workers are widely distributed across the subsidy brackets, it makes sense to consider the role of the firm in affecting their subsidy rates through influencing job duration. Figure 1 illustrated that a firm can benefit greatly from retaining the worker a bit longer if she is just below the 120- or 400-h threshold. For example, at $6 per hour, a worker at the 399-h mark qualifies the firm for about $600, whereas the same worker who remains for the 400th hour qualifies the firm for almost $1,000 (a 60% increase in benefits). Firms can even benefit by keeping a worker for a substantial amount of extra time to reach the threshold: if this worker had been at the firm for only 340 h, the firm would still be better off financially if it retained the worker until the 400th hour. (26) This calculation does not even include the added value of any production accomplished by the worker in the extra 60 h of work.

It is clear that there are strong incentives for firms to influence job duration of their WOTC workers, especially because the costs of participating in the program are sunk. However, there is not much evidence of overall improvements in average job duration (measured in quarters of work) due to WOTC participation (Hamersma 2008). It is possible, though, that even without increasing average job duration substantially firms may still react to the incentives to increase tenure near the thresholds, where the marginal subsidy benefits are largest.

I test for this behavior using an estimate of the hours distribution of the whole population of WOTC workers in Wisconsin during 1999-2002. If firms increase job duration near the hours thresholds, there should be a dip in the distribution just below these thresholds and a spike just above them, indicating that people who would have left with job durations just below the thresholds were kept for the extra hours needed to receive the increased subsidy. I use a kernel density estimator to check the distribution for these features. Saez (2002) estimates kernel densities to assess responses to marginal tax rates, for which he finds little evidence. However, using a similar technique, Blundell and Hoynes (2004) find evidence of substantial responses to an hours-worked requirement for a U.K. welfare program.

The estimated hours distribution for WOTC workers is shown in Figure 4. The figure presents the estimated hours for workers with job durations between 50 and 600 h. These data are only slightly smoothed from their original form; this undersmoothing allows the best possibility of observing even a small irregularity near the threshold, although it also displays a great deal of the noise in the data. Note also that there is some measurement error in hours worked (due to my imperfect measure of wages) which could smooth the distribution a bit, so while we want to focus on irregularities near each threshold, we should look further than just a few hours on each side. (27) The two vertical lines in the figure mark the WOTC thresholds. On the basis of the figure, there is no irregularity near the 120-h threshold, but potentially a small dip-and-spike pattern near the 400-h threshold.

[FIGURE 4 OMITTED]

To look more closely at the potential irregularity near the 400-h threshold, I plot a narrower slice of this distribution in Figure 5, examining only those with tenure between 320 and 480 h. This "closeup" seems to provide some support for the hypothesis that firms may be retaining workers up to the 400-h mark who would otherwise have left 10-20 h earlier. However, the pattern is weak enough that this graph is also consistent with no response by employers; a similar small dip-and-spike pattern also occurs in other parts of the distribution where no relevant thresholds are crossed, so it may merely reflect random fluctuations. In fact, if the graph is smoothed much, the pattern near the 400-h threshold disappears. However, measurement error in hours could also potentially contribute to my failure to find a distinct dip-and-spike pattern. Unfortunately, this is the best measure of hours available, so this concern remains a limitation of the analysis. I can simply conclude that I find no strong evidence of the dip-and-spike pattern that would be predicted by economic theory.

This analysis of the distribution of WOTC workers' hours of work does not suggest substantial intervention by employers to maximize credits. There are several possible reasons for this result. First, firms may not have the ability to influence job duration on the margin, perhaps because many of the job separations in this population are quits. Second, managers may not be constantly aware of the number of hours worked by their WOTC workers, so they may not be able to take this into account when making decisions about scheduling or firing a marginal worker. Third, even if hours are known, firms may not educate managers on the hours-related incentives created by the WOTC program. Finally, managers may be unaware of which workers are WOTC certified. This seems particularly likely if the subsidy program is administered through an outside tax consultant, which is the case for most WOTC-participating firms in Wisconsin.

[FIGURE 5 OMITTED]

Although I cannot easily disentangle these possible explanations, I can explore the first by examining worker quit rates at the industry level to see whether they are correlated with WOTC participation. Among the four industries that encompass most of the workers in my sample, Table 4 suggested that "Health Care and Social Assistance" and "Administrative" (mostly THS) firms were the most unlikely to claim the credits, all else equal. If workers tend to quit more in those industries, there is little opportunity for firms to capitalize on adjusting hours to cross the WOTC thresholds; this may be a reason for low participation. Table 5 provides industry-level data on turnover and quit rates from the Bureau of Labor Statistics (BLS) Job Openings and Labor Turnover Survey (JOLTS).

The statistics in Table 5 do not suggest that the low-participation industries have particularly high quit rates relative to the others. (28) In fact, the two industries with higher participation ("Accommodation and Food Service" and "Retail Trade") have very high quit rates as a percentage of separations (70% and 64%, respectively) and high rates of separation. "Health Care and Social Assistance" has a similar rate of 68%. By comparison, "Professional and Business" has a rate of only 55%. Rather than responding to quit rates, WOTC participation patterns may instead be responding to a different facet of employment patterns: "Accommodation and Food Service" and "Retail Trade" have higher hiring rates each month than the other industries. The constant stream of new hires (and thus potential WOTC subsidies) may be one reason these industries participate at a higher rate, even though many workers quit and thus prevent the firm from maximizing subsidy receipts.

Given these interesting differences across the key industries represented in my data, it is also useful to check whether industries are homogeneous in their apparent lack of response at the hours thresholds. Figure 6 provides separate earnings distributions by industry, and suggests no systematic pattern. It is possible that there is some response to the subsidy at the 120-h threshold in "Health Care and Social Assistance" (although the dip is a bit early) and possibly a small response at the 400-h threshold in "Accommodation and Food Service." However, because these industries have high quit rates, it seems unlikely that the response from firms could be large. Ultimately, the response to the hours threshold appears weak and unsystematic across industries if it exists at all.

Because firms do not respond significantly to WOTC hours thresholds, it seems unlikely that the hours distribution of workers in a given firm is strongly influenced by the firm's WOTC participation. If endogeneity is present, it must be that firms increase durations more broadly in response to the WOTC but ignore the thresholds, which is not consistent with the findings in Hamersma (2008) or with maximizing subsidy gains. (29) Firms' lack of response at the thresholds suggests that the hours distribution was not endogenous in the participation equation estimated in the previous section. On the basis of this finding, I conclude that the positive relationship found between job duration and WOTC participation primarily reflects a causal link going in only one direction: firms' job duration distributions are a factor ex ante in their decisions about WOTC participation.

VIII. CONCLUSION

This study provides the first available evidence on the relationship between workers' job duration distributions and firms' willingness to participate in employer subsidy programs with hours requirements. In comparison with firms that participate in the WOTC subsidy program, I find that eligible nonparticipating firms tend to have fewer workers who have surpassed the 120- and 400-h thresholds that define increases in the WOTC subsidy rate. I estimate a participation equation at the firm level and find that, even controlling for the number of eligible workers, firms are more likely to participate if a larger percentage of their workers are in the higher subsidy brackets. I investigate whether the job durations in this equation (reflected in the subsidy brackets) might be endogenous, because their correlation to participation may reflect firms' attempts to improve job durations after choosing to participate. There is very little evidence of firms responding to the subsidy incentives by trying to influence job duration. Specifically, participating firms do not appear to respond to the strongest job duration incentives of the WOTC, which occur near the hours thresholds and could result in large increases in the subsidy for small increases in tenure. Because firms do not appear to respond to these crucial points, this suggests that job duration is exogenous in the participation equation. I conclude that workers' job duration distribution has a causal impact on firms' participation in the WOTC, most likely because of its significant effect on potential benefits due to widely varying subsidy rates created by the program's hours requirements.

If policy makers hope to increase participation in the program, they may need to consider the potentially discouraging role of hours requirements. Note that there are no such requirements in the key federal worker subsidy program--the Earned Income Tax Credit--and it enjoys high participation. However, there were important reasons for these requirements in the original design of the WOTC: policy makers hesitated to encourage employment that does not last very long and thus may not contribute to long-run improvements in the labor market outcomes of the disadvantaged. (30) Both these considerations will be relevant to the discussion as Congress considers reauthorization of WOTC in the coming years.

[FIGURE 6 OMITTED]

ABBREVIATIONS

CARES: Client Assistance for Re-employment and Economic Support

DOL: Department of Labor

ETA: Employment and Training Administration

GAO: General Accounting Office

IRP: Institute for Research on Poverty

IRS: Internal Revenue Service

JOLTS: Job Openings and Labor Turnover Survey

OIG: Office of the Inspector General

SSN: Social Security Number

THS: Temporary Help Services

TJTC: Targeted Jobs Tax Credit

UI: Unemployment Insurance

WOTC: Work Opportunity Tax Credit

WtW: Welfare-to-Work Tax Credit

APPENDIX

Description Data Sources:

Wisconsin WOTC/WtW database

Who: Universe of all applications for WOTC from employers during 1999-2002

What: masked social security number

employer ID

target group (welfare, food stamps, ex-felon, etc.)

starting wage

job start date

occupation

certification date (if application approved)

Purpose: can identify WOTC/WtW certified workers and their employment information

Wisconsin CARES data

"Client Assistance for Re-employment and Economic Support"

Who: Universe of all WI recipients of welfare or food stamps during calendar years 1998-2001

What: masked social security number

monthly welfare and food stamp records

demographics 

Purpose: can identify subset of workers who are WOTC eligible via welfare or food stamps

can merge with WOTC/WtW database to identify which eligible workers (via welfare or food stamps) were WOTC-certified

Wisconsin Unemployment Insurance database

Who: Everyone in the CARES and WOTC data; quarterly person-job records, 1995-2004

What: masked social security number

employer ID

quarterly earnings (by job)

industry

Purpose: can merge with both WOTC and CARES info to examine labor market outcomes

[FIGURE A1 OMITTED]

TABLE A1
Results of Log Wage Regression
Dependent Variable: In (wage)

Variable                      Coefficient     t-statistic

Demographics:
Female                            -0.010         -0.99
Black                             -0.013         -1.59
Hispanic                          -0.013         -0.95
Other race (non-white)            -0.013         -1.41
Age                                0.011          3.12
Age squared                       -0.0002        -3.09
GED                               -0.003         -0.29
High school diploma                0.011          1.87
Some college                      -0.002         -0.14

Firm characteristics:
County avg. # workers per          0.001          1.22
  firm (by industry)
Firm headquarters in WI           -0.090        -15.59
  (1 = yes, 0 = no)

Industry indicators:
Administrative and support         0.182         12.12
  and waste management
Finance and insurance              0.517          3.93
Health care and social             0.299         20.10
  assistance
Information                        0.045          0.86
Management of companies            0.400          5.29
  and enterprises
Manufacturing                      0.167          2.56
Real estate and rental and        -0.050         -1.33
  leasing
Retail trade                       0.051          6.13
Transportation and                 0.271         13.13
  warehousing
Wholesale trade                   -0.073         -2.35

Notes: N = 2,535 worker-jobs. Adjusted [R.sup.2] = 0.4040.
There are 2,621 worker-jobs with wage data, but 86 are
dropped from the regression due to missing data on region,
education, or industry. These 86 are retained throughout
the following analysis, though, because the purpose of this
regression is to generate imputations (their wage data do
not need to be imputed, since they are directly available).
The omitted industry category is "Accommodation and Food
Services," the omitted educational category is "Less than
high school," and the omitted race is "White." Region and
quarter dummies are also included but not reported.

REFERENCES

Bishop, J. H., and S. Kang. "Applying for Entitlements: Employers and the Targeted Jobs Tax Credit." Journal of Policy Analysis and Management, 10(Winter), 1991, 24-45.

Blundell, R., and H. Hoynes. "Has 'In-Work' Benefit Reform Helped the Labour Market?" in Seeking a Premier League Economy, edited by R. Blundell, D. Card, and R. Freeman. Chicago: University of Chicago Press, 2004.

Burtless, G. "Are Targeted Wage Subsidies Harmful? Evidence from a Wage Voucher Experiment." Industrial and Labor Relations Review, 39, 1985, 105-14.

Christensen, S. The Targeted Jobs Tax Credit. Staff Memorandum. Washington, DC: Congressional Budget Office, May 1984.

Dickert-Conlin, S., and D. Holtz-Eakin. "Employee-Based versus Employer-Based Subsidies to Low-wage Workers: A Public Finance Perspective," in Finding Jobs: Work and Welfare Reform, edited by D. Card and R. Blank. New York: Russell Sage Foundation, 2000, 262-95.

Gunderson, J. M., and J. L. Hotchkiss. "Job Separation Behavior of WOTC Workers: Results from a Unique Case Study." Social Service Review, June 2007, 317-42.

Hamersma. S. "An Evaluation of the Work Opportunity Tax Credit: Participation Rates and Employment Effects." National Tax Journal, 56, 2003, 725-38.

--. "The Effects of an Employer Subsidy on Employment Outcomes: A Study of the Work Opportunity and Welfare to Work Tax Credits." Journal of Policy Analysis and Management, 27(3), 2008, 498-520.

Hamersma, S., and C. Heinrich. "Temporary Help Service Firms' Use of Employer Tax Credits: Implications for Disadvantaged Workers' Labor Market Outcomes." Southern Economic Journal, 74(4), 2008, 1123-48.

Katz, L.F. "'Wage Subsidies for the Disadvantaged," in Generating Jobs: How to Increase Demand for Less-Skilled Workers, edited by R. B. Freeman and P. Gottschalk. New York: Russell Sage Foundation, 1998.

Lorenz, E. C. "TJTC and the Promise and Reality of Redistributive Vouchering and Tax Credit Policy." Journal of Policy Analysis and Management, 14(Spring), 1995, 270-90.

O'Neill, D. M. "Employment Tax Credit Programs: The Effects of Socioeconomic Targeting Provisions." Journal of Human Resources, 17(Summer), 1982, 449-59.

Richer, E., S. Savner, and M. Greenberg. Frequently Asked Questions about Working Welfare Leavers. Washington, DC: Center for Law and Social Policy, 2001.

Saez, E. "Do Taxpayers Bunch at Kink Points?" National Bureau of Economic Research Working Paper No. 7366, 1999. Updated April 2002.

Salamon, L. M., and S. W. Sokolowski. "Nonprofit Organizations: New Insights from QCEW Data." Monthly Labor Review, September 2005, 19-26.

U.S. Department of Labor. Office of the Inspector General (OIG). Targeted Jobs Tax Credit: Employment Inducement or Employer Windfall? Washington, DC: U.S. Department of Labor. Office of the Inspector General, 1994.

U.S. General Accounting Office (GAO). Work Opportunity Tax Credit: Employers Do Not Appear to Dismiss Employees to Increase Tax Credits. Washington, DC: U.S. General Accounting Office, 2001, GAO-1-329.

(1.) Gunderson and Hotchkiss (2007) report measures of job duration in days for a subset of WOTC employees within one large company.

(2.) See details at http://www.netprofitincrease.com/services_wotc.asp.

(3.) The WOTC was initially introduced as part of the Small Business Job Protection Act of 1996 (P.L. 104-188). The WtW was introduced as part of the Taxpayer Relief Act of 1997 (P.L. 105-34). Both programs have been repeatedly reauthorized, generally for 2 yr at a time. They were recently reauthorized in the Tax-Relief and Health Care Act of 2006 (P.L. 109 432), at which time they were combined into a single program and some target group definitions were modified (including a wider age range for food stamp recipients). The program has since been further extended to workers hired anytime before August 31, 2011 (P.L. 110-28).

(4.) A study by the General Accounting Office (GAO) (2001) suggests that only a very small portion of WOTC workers reach the maximum credit, and that firms do not appear to dismiss these workers when they reach this maximum credit (as we might expect based on the incentives of the program). I can confirm that a similarly small portion of my sample achieves the maximum credit: given the small size of this group and the evidence from GAO (2001), l do not spend time addressing this aspect of the program's incentives in this study.

(5.) The WtW is only potentially more valuable to the firm than the WOTC in the first year if a worker has enough earnings to reach the maximum WOTC subsidy ($6,000 in earnings, for the maximum $2,400 credit/ and continues to generate additional earnings. This is because the WtW maximum credit is higher, at $3,500 (applying to up to $10,000 in earnings), even though its subsidy rate is lower (35% rather than 40%). The average government spending per WtW certification granted is only about $325 (Hamersma 2003), which suggests that the vast majority of WtW certifications do not get claimed under WtW, but are instead claimed under WOTC or not at all.

(6.) The program also allowed "180 d of employment" as an alternative to the 400-h requirement. I locus on the hours requirements rather than the "number of days" requirements. This is primarily because the hours requirement would in most cases be met before the days requirement (making the days requirement nonbinding) and also because my data allow me to estimate hours of work but not number of days employed.

(7.) On the basis of my microdata from Wisconsin, the vast majority of WOTC applications in 1999-2002 were submitted by tax consultants on behalf of firms. Consultants in some cases handle all documentation and communication with the worker, such that local staff are not even aware of who has been certified (Hamersma and Heinrich 2008).

(8.) The relevant forms are IRS 5884 for the WOTC and IRS 8861 for the WtW. Firms are not required to provide documentation of hours worked unless requested in an audit.

(9.) All these data were generously provided by the Wisconsin Department of Workforce Development, in conjunction with the University of Wisconsin's Institute for Research on Poverty (IRP). The employer identifiers in the U1 data are based on a different system than those in the WOTC data, so I obtained a "matching" file for these identifiers from IRP that allows me to link the appropriate job records.

(10.) The fraction of food stamp youth (21%) is consistent with their fraction of WOTC certifications nationally (Hamersma 2003), and the traction of welfare recipients is on the lower end in Wisconsin (35%) relative to national patterns (30%-60%) (Gunderson and Hotchkiss 2007).

(11.) I drop those that are coded as "less than the minimum wage," because l do not know their wage range and most are in the service industry, suggesting that they may work for tips which I cannot observe. For those who report wages "$9.00 or greater," I impute a wage of $10 per hour. Estimates are not sensitive to the chosen imputation (even a much higher imputation of $12 per hour results in less than a 0.5% change in the size of each category).

(12.) An informal survey of local branches of some large firms in the retail and accommodations industries indicated that individuals seldom qualify for a raise prior to 90 days of employment. These firms were generally following a company-wide policy, so geographic variation should be minimal.

(13.) However, it is noteworthy that within the same job description, non-WOTC workers had a similar separation rate (Gunderson and Hotchkiss 2007).

(14.) The welfare and food stamp target groups are among the largest of the WOTC target groups (whereas the WtW includes only welfare recipients). The welfare target group consists of those with at least 9 mo of welfare receipt in the 18 mo prior to their job start. The food stamp group consists of workers aged 18-24 who have received food stamps for at least the Past 6 mo (or at least 3 of the last 5 mo but are now ineligible).

(15.) Some of the workers' wages cannot be imputed this way. Some of them, for instance, work in industries that do not employ anyone for whom I have wage data. Out of 32,381 worker-jobs without wage data, there are 1,901 for whom I cannot impute wages for this reason. Similarly, I drop an additional 33 worker-jobs without wage data who have college degrees (or more) because no worker with wage data has a college degree. In addition, 1.370 worker-jobs without wage data are missing at least one variable in the regression, so I am also unable to impute wages for them and they are dropped. In total, 1 drop 3,304 worker-jobs, resulting in a sample of 29,077 worker-jobs with imputed wages.

(16.) Hamersma (2008) reports an average WOTC earnings premium of about 10%.

(17.) For example, some of the larger firms requesting the WOTC in Wisconsin asked the WOTC office to stop sending paper rejections for all of their (many) applications on behalf of unqualified workers; this suggests that some employers do not limit WOTC paperwork to those anticipated to be eligible. In addition, some large employers that post application packets online can be seen to include the WOTC prescreening notice in the packet of materials.

(18.) Unfortunately, the data available are not sufficiently rich to sensibly model this partial participation. For this reason, I focus on the binary decision to participate, even though it is in reality followed by some (apparently more complicated) decision regarding "how much" to participate.

(19.) Because the sample covers a 2.5 yr span of time, some firms start participating partway through the sample period. They may mistakenly appear to participate at a very low level if I divide total applications by total eligible workers during the sample period. Instead, I evaluate participation relative to the number of eligible workers after the firm begins participating. Out of 663 firms in the sample with at least ten eligible workers in July 1999 to December 2001, there are 506 with no participation (19,589 workers), 43 with less than 20% participation (3,227 workers), and 114 with greater than 20% participation (8,882 workers).

(20.) It is possible that some of the nonparticipants are not eligible due to being public, being nonprofit, or having no tax liability, which I cannot determine given my data. However, this should be a small portion of the sample, as only about 9% of private employment is in the nonprofit sector in the East North Central census region (IL, IN, MI, OH, and WI). Nationally, almost 90% of these jobs are in the health care, social services, or education sectors (Salamon and Sokolowski 2005).

(21.) This support for comparative statics without a full explanation for the overall level of participation runs parallel to some key results in the voting literature: the rate of voter participation is higher than would be predicted by theory, but the comparative static predictions regarding expected influences on voter turnout (such as weather) are consistently borne out in the data.

(22.) One consulting company, TaxBreak, states: "TaxBreak works for you on a contingency fee basis. If you do not recover monies through our tax credit and recovery services, we do not get paid." (http://www. psapayroll.com/taxbreak_home.html). David Loney, a tax consultant who was convicted of inflating tax credit claims, also received "a commission based on the dollar value of the WOTC and WtW tax credits he obtained for his clients" (http://www.oig.dol.gov/public/media/dloney.html).

(23.) The 95% confidence interval is ($126,000, $164,000). Because the firm must decide to participate ex ante, and I measure potential benefits ex post (by using the total hours worked by each employee to assign subsidy rates appropriately), this participation equation does not account well for risk. In an effort to consider the greater uncertainty (in dollar terms) associated with certifying many workers rather than just a few, l also estimate this equation including the square of the potential benefit ([B.sup.2.sub.i]) in the probit. The prediction is that participation would be decreasing in [B.sup.2.sub.i], although still increasing on net with Bi. My findings are consistent with this, but the estimated fixed costs change only slightly, to about $130,000.

(24.) This coefficient estimate is statistically significantly larger than both those for the shorter duration (120-299 h) and the longest duration (maximum subsidy) jobs. It is not clear why the longest duration jobs do not seem to have the largest effect. Part of the issue may be that after workers reach the maximum subsidy, firms no longer have incentives to further extend job duration. In addition, long job durations may indicate that the firm has little turnover and thus encounters fewer opportunities to apply for the subsidy overall, which could counteract the incentive to participate.

(25.) Detailed tables are available upon request. Statistical significance here is defined as 90% confidence or more.

(26.) For the 340-h case, the firm's wage costs are $6 x 340 = $2,040, with a subsidy of $2,040 x .25 = $510, for net costs of $1,530. For the 400-h case, wages are $2,400 with a subsidy of $960, for net costs of $1,440.

(27.) For example, consider a worker in the modal wage category ($6.00, $6.99). This worker would be estimated to work for exactly 120 h (the first threshold) if her total earnings were $780 ($6.50 x 120). If she actually earned $6.00 per hour, her true hours were 130; if she actually earned $6.99, her true hours were 112. This means that my use of the midpoint of the wage category--at least for those in the area near the first threshold--should result in estimates within about 10 h of the true value.

(28.) Unfortunately, the "'Administrative" category was aggregated in the JOLTS data to include other business-related industries so this may not be a fully accurate representation of patterns in the "Administrative" industry specifically.

(29.) Hamersma (2008) found job durations in Wisconsin, in quarters, to be basically unaffected by the WOTC.

(30.) In fact, evidence regarding the effects of the subsidy on worker outcomes (such as wages) suggests that the program generates only limited improvements even when firms do participate (Hamersma 2008).

Hamersma: Department of Economics, University of Florida, Gainesville, FL 32611-7140. Phone [phone], Fax [phone], E-mail sarah.hamersma@ cba.ufl.edu

doi:10.1111/j.1465-7295.2009.00260.x

TABLE 1
Descriptive Statistics for WOTC-Certified
Workers in Wisconsin, 1999-2002

                                         Percent in
                                          Category

Target group                             N = 17,048
  WtW (or both WOTC and WtW)                24.8
  Food stamp youth (18-24)                  21.1
  High risk youth (18-24, living            13.5
   in empowerment zone/enterprise
   community)
  Welfare (WOTC only)                        9.9
  Ex-felon                                   9.5
  Vocational rehabilitation                  9.2
  SSI                                        9.0
  Veteran                                    1.6
  Summer youth                               1.4
Starting wage                            N = 16,958
  <Minimum wage                              2.3
  $5.15-$5.99                               24.1
  $6.00-$6.99                               35.7
  $7.00-$7.99                               17.2
  $8.00-$8.99                               12.7
  $9.00 and up                               8.1
Occupational category
(WOTC record)                            N = 16,976
  Service                                   35.3
  Clerical and Sales                        32.8
  Professional/Technical/                    9.5
   Managerial
  Machine trades                             1.4
  Processing                                 0.7
  Structural                                 0.6
  Farm/Forestry/Fishery                      0.6
  Other                                     19.2
Industry of firm (UI record)             N = 16,891
  Retail Trade                              38.5
  Accommodation/Food Services               20.0
  Administrative/Support/Waste              19.7
Management
  Health Care/Social Assistance              8.7
  Manufacturing                              5.0
  Transportation/Warehousing                 3.5
  Other                                      4.6

Notes: This sample contains 17,281 observations. The
total number of worker-jobs in the WOTC records for this
time period is 20,577. Of these, 725 (about 3.5%) did not
have any UI records in 1999-2004 (including 27 whose
records had not been requested from UI), and about 13%
of the remainder did not successfully match with a UI
record for the WOTC job (either there was no UI record
indicating the employer listed in the WOTC record or no
record in the appropriate quarter that could be identified as
the WOTC job). The remaining 17,281 observations reflect
the records of 14,181 distinct individuals, some of whom had
more than one WOTC record during 1999-2002. Separate
sample sizes are given for each variable because some
records have missing data for some variables. I do not use
the available partial-year WOTC data from 1998 or 2003
because it is incomplete. Wage categories are nominal. SSI
indicates recipients of Supplementary Security Income. a
program for individuals with disabilities. Sample: all WOTC
certifications in Wisconsin for jobs starting in 1999-2002.

TABLE 2
Fraction of Firms that Participate in WOTC, by Firm's Potential
Tax Credit

Potential          Less than   $5,000-   $10,000-   $20,000-
Credit                $5,000   $10,000    $20,000   $30,0(10

Number of firms           85       124        179         98
Fraction
Participants             .14       .13        .12      16.00
Nonparticipants          .86       .87        .88        .84

Potential          $30,000-   $50,000-   $100,000-   $200,000
Credit              $50,000   $100,000    $200,000   or more

Number of firms          63         48          16          7
Fraction
Participants            .32        .33         .56        .57
Nonparticipants         .68        .67         .44        .43

Notes: N = 620 firms (506 nonparticipants and 114 participants).
All firms in sample have at least ten eligible workers during
July 1999 to December 2001. Firms that applied for the WOTC at
some point but did so for fewer than 20% of their eligible
workers after that point are excluded. The intervals listed are
closed on the left and open on the right.

TABLE 3
Results of Participation Probit for Estimating
Fixed Costs

                                     Standard
Variable               Coefficient    Error     z-Statistic

Expected tax               0.00772    0.00158          4.88
  credit for firm
  (in $1,000s)
Constant term              -1.1220    0.07512        -14.94

Notes: See sample description in Table 2. Pseudo
[R.sup.2] = .045.

TABLE 4
Results of Participation Probit with Hours Distribution

Variable                                  Marginal Effect   z-Statistic

Percent of eligible workers with               .0025          1.88
  120-399 h work
Percent of eligible workers with               .0056          3.52
  400+ h work, not max
Percent of eligible workers                    .0013          1.12
  qualifying for maximum subsidy
Number of eligible workers at firm             .0004          1.94
Average wage paid by firm to                   .1446          2.69
  sample workers
Industry indicators: (Accommodation/
  Food Service omitted)
Administrative/Support/Waste                  -.1621         -2.47
  Management (THS)
Retail Trade                                  -.0185         -0.37
Health Care and Social Assistance             -.2431         -3.14
Information                                    .0257          0.20
Management of Companies and Enterprises       -.1047         -0.61
Manufacturing                                 -.1419         -2.29
Real Estate and Rental and Leasing             .3511          1.27
Finance and Insurance                         -.1651         -2.60
Transportation and Warehousing                -.1411         -2.01
Wholesale Trade                               -.1045         -1.17

Variable                                   Sample Mean

Percent of eligible workers with            25.0
  120-399 h work
Percent of eligible workers with            13.3
  400+ h work, not max
Percent of eligible workers                 16.8
  qualifying for maximum subsidy
Number of eligible workers at firm          43.7
Average wage paid by firm to                 7.19
  sample workers
Industry indicators: (Accommodation/
  Food Service omitted)
Administrative/Support/Waste                  .272
  Management (THS)
Retail Trade                                  .151
Health Care and Social Assistance             .235
Information                                   .010
Management of Companies and Enterprises       .003
Manufacturing                                 .049
Real Estate and Rental and Leasing            .003
Finance and Insurance                         .026
Transportation and Warehousing                .018
Wholesale Trade                               .008

Notes: N = 617 firms. The reported marginal effects are the effects
at the mean of the covariates. The three additional firms that were in
Table 3 had either missing regressors or no variation in participation
within their region or industry. Pseudo [R.sup.2] = .172. Omitted
categories: percent of workers with < 120 h of work and "Accommodation
and Food Services" industry. Regional indicators, as well the number
of counties represented in the firm's workforce (a measure of
dispersion), are also included in the estimation. Firm is labeled as
a nonparticipant if none of its eligible hires are claimed and as a
participant if it submits applications for at least 20% of its
eligible hires after its first application. (Those firms with
positive participation below 20% are not included in the sample.)
The fraction of participating firms in this sample is 18.2%.
Standard errors are robust to heteroskedasticity but do not account
for possible measurement error in assignment to job duration
categories (which should be minimal given the broadness of the four
categories) or the imputation of wages for some workers. Dependent
variable: firm participates in WOTC.

TABLE 5
Employment Patterns by Industry, 2001

                                   Average Monthly    Average Monthly
Industry                           Hiring Rate (%)  Separation Rate (%)

Health Care and Social Assistance        3.04               2.73
Professional and Business (THS +)        4.31               3.98
Retail Trade                             4.72               4.78
Accommodation and Food Service           7.23               6.99

                                    Average Quits as Percent
Industry                            of Average Separations

Health Care and Social Assistance            67.58
Professional and Business (THS +)            54.72
Retail Trade                                 64.05
Accommodation and Food Service               70.44

Note: The category "Professional and Business" reported by Bureau
of Labor Statistics (BLS) combines the "Administrative" industry
used in my analysis with the "Management of Companies and
Enterprises" and "Professional, Scientific, and Technical
Services" industries. The hiring and separation rates are a
fraction of total employment. Data source: BLS Job Openings and
Labor Turnover Survey (JOLTS).